The carbon balance of European croplands: A cross-site comparison of simulation models

Croplands cover approximately 45% of Europe and play an important role in the overall carbon budget of the continent. However, the estimation of their carbon balance remains uncertain due to the diversity of crops and cropping systems together with the strong influence of human management. Here, we present a multi-site model comparison for four cropland ecosystem models namely the DNDC, ORCHIDEE-STICS, CERES-EGC and SPA models. We compare the accuracy of the models in predicting net ecosystem exchange (NEE), gross primary production (GPP), ecosystem respiration (Reco) as well as actual evapo-transpiration (ETa) for winter wheat (Triticum aestivum L.) and maize (Zea mays L.) derived from eddy covariance measurements on five sites along a gradient of climatic conditions from eastern to south-westerly Europe. The models are all able to simulate daily GPP. The simulation results for daily ETa and Reco are, however, less accurate. The resulting simulation of daily NEE is adequate except in some cases where models fail due to a lack in phase and amplitude alignment. ORCHIDEE-STICS and SPA show the best performance. Nevertheless, they are not able to simulate full crop rotations or the multiple management practices used. CERES-EGC, and especially DNDC, although exhibiting a lower level of model accuracy, are able to simulate such conditions, resulting in more accurate simulation of annual cumulative NEE.

Research highlights▶ We present a comprehensive model comparison, of the most prominent model species used to simulate ecosystems, from semi-empirical to process oriented, form site to global vegetation model, using eddy-flux data for carbon and water from European croplands sites. ▶ The paper is able to identify possible inconsistencies between methods used to process eddy-flux data, e.g. u* correction, gapfilling, etc., and the process description in the models. We discuss the possible reasons for the poor performance of all models in reproducing low carbon fluxes ranging −2 g C m2 day−1 and 2 g C m2 day−1 in a comprehensive way, highlighting possible sources of disagreement. ▶ We conclude that the accurate representation of crop phenology, crop management and post harvest events is a prerequisite to produce accurate annual carbon fluxes. ▶ A relatively poor performance in reproducing the latent heat fluxes for DNDC and CERES-EGC, especially under warm and dry conditions, in connection with a satisfying performance in carbon flux simulations, suggesting problems in the coupling of the water and carbon flux in these models.